无线传感器网络异常值检测研究

发布时间：2018-04-04 06:07

本文选题：无线传感器网络　切入点：异常值检测　出处：《西安电子科技大学》2014年硕士论文

【摘要】：无论是以事件监测还是时空数据采集为目的的无线传感器网络,都迫切需要一种实时的异常值检测方法。然而,无线传感器网络节点的计算能力、存储能力和电源电量都非常有限。这为寻找适用无线传感器网络的异常值检测方法造成了很多限制。本文针对无线传感器网络中数据具有时空相关性特点提出了两种适用于无线传感器网络的异常值检测方法。基于贝叶斯分类器的异常值检测方法。首先,在超椭球几何理论的基础上给出一种节点相似度的定义。当网络部署在复杂环境中时,利用本文定义的相似度选择邻居节点,优于仅仅依靠节点间距离远近的方法。其次,对网络进行分簇并为簇内构建链式的贝叶斯分类器进行推理,在提高检测准确率的同时,将可能的错误传播限制在一定范围,避免对整个网络的破坏。该模型可以实现实时的异常值检测。基于可变阈值的异常值检测方法。从数据的时间相关性出发,定义了一种异常因子,并通过异常因子将数据分为正常态,临界态和异常态。计算异常因子公式的分母是一个随数据动态变化的阈值。在马尔科夫假设和大量仿真实验的基础上,结合数据的三种分类给出了动态阈值随数据变化的更新机制。该模型可以实现实时的异常值检测。
[Abstract]:A real-time outlier detection method is urgently needed in wireless sensor networks (WSN) for event monitoring and spatio-temporal data acquisition.However, the computing power, storage capacity and power supply of wireless sensor network nodes are very limited.This creates a lot of limitations for finding outlier detection methods suitable for wireless sensor networks.In this paper, two outliers detection methods for wireless sensor networks (WSN) are proposed in view of the spatio-temporal correlation of data in wireless sensor networks (WSN).An outlier detection method based on Bayesian classifier.Firstly, a definition of node similarity is given based on hyperellipsoid geometry theory.When the network is deployed in a complex environment, it is better to select neighbor nodes by using the similarity defined in this paper than by only relying on the distance between nodes.Secondly, the network is clustered and a chain Bayesian classifier is constructed within the cluster to improve the detection accuracy and limit the possible error propagation to a certain range to avoid the destruction of the whole network.The model can detect outliers in real time.Outlier detection method based on variable threshold.Based on the temporal correlation of data, an anomaly factor is defined, and the data is divided into normal state, critical state and abnormal state.Calculating the denominator of the anomaly factor formula is a threshold value that varies dynamically with the data.On the basis of Markov hypothesis and a large number of simulation experiments, the updating mechanism of dynamic threshold with the change of data is given in combination with three kinds of classification of data.The model can detect outliers in real time.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP212.9;TN929.5

【相似文献】